BREAKING CHANGE: browser.evaluateEnabled now defaults to false. The evaluate feature allows arbitrary JavaScript execution in the browser context via the /act endpoint (kind=evaluate) and wait --fn. This is a powerful escape hatch for advanced automation but poses security risks: - Prompt injection: LLM agents can be tricked into running malicious code - Network exposure: If gateway is accessible, attackers can exfiltrate data - Session hijacking: Code runs with access to cookies, localStorage, DOM With this change: - act:evaluate and wait --fn return 403 by default - Users who need this feature can enable it explicitly: browser.evaluateEnabled: true Built-in actions (click, type, screenshot, wait for text/URL, etc.) remain fully functional and cover the vast majority of automation use cases.
9 lines
451 B
TypeScript
9 lines
451 B
TypeScript
export const DEFAULT_CLAWD_BROWSER_ENABLED = true;
|
|
export const DEFAULT_BROWSER_EVALUATE_ENABLED = false;
|
|
export const DEFAULT_CLAWD_BROWSER_COLOR = "#FF4500";
|
|
export const DEFAULT_CLAWD_BROWSER_PROFILE_NAME = "clawd";
|
|
export const DEFAULT_BROWSER_DEFAULT_PROFILE_NAME = "chrome";
|
|
export const DEFAULT_AI_SNAPSHOT_MAX_CHARS = 80_000;
|
|
export const DEFAULT_AI_SNAPSHOT_EFFICIENT_MAX_CHARS = 10_000;
|
|
export const DEFAULT_AI_SNAPSHOT_EFFICIENT_DEPTH = 6;
|